Art, technology & authorship · ZGenArt.com editorial · October 24, 2026

A PNG file is not a confession about how an artwork was made. It may contain an image of a watercolor on cotton paper, a photograph of an oil-pastel painting, a drawing made by hand with a stylus, a hybrid composition, or an image generated by a model. The filename extension does not settle the question of authorship.

This distinction matters because digital platforms increasingly attach AI-related labels to pictures. Artists deserve transparency, but transparency that collapses very different processes into one category can mislead the public about the artwork and the person who made it.

The file is not the medium

Consider a painter who creates an original work using physical pigment on paper. A camera records the painting, the image is color-corrected for faithful presentation, and the final file is exported as PNG. Nothing in that sequence turns the original painting into AI-generated art. The PNG is a means of displaying the work, not a replacement for the work’s physical medium.

The same principle applies to an artist drawing directly in Infinite Painter or another digital painting program. A human hand can create every mark on a tablet. Digital painting is not synonymous with generative AI. The use of software alone tells us very little about who chose the forms, placed the marks or developed the composition.

Four processes that should not be confused

Photographed physical art: a traditional painting, drawing or sculpture reproduced as an image file.

Human-made digital painting: artwork constructed by an artist using digital brushes, layers and a stylus.

Hybrid practice: work combining physical, digital and sometimes AI-assisted steps, with the actual contribution of each described accurately.

Generative AI imagery: work in which a model generates visual material from instructions or other inputs. The degree of human direction, selection and subsequent alteration may still vary.

The border between these practices is often more informative than a simple yes-or-no badge. Proper attribution should describe the production of the particular artwork instead of relying on assumptions about what its pixels resemble.

What automated labels can and cannot establish

Technical signals may help a platform infer that some kind of AI tool was involved. They are not an infallible, universally applicable test of artistic origin. Classifiers can make mistakes; relevant metadata can be missing or transformed during an editing and export workflow; and different tools can leave different traces.

Meta itself acknowledged that some images involving only minor AI edits received a broad “Made with AI” label and changed the wording to “AI info.” In its published explanation, Meta distinguished content it regarded as AI-generated from content merely modified with AI tools. That distinction is precisely why the scope and meaning of a label deserve scrutiny.

The Meta explanation of AI labeling is an important starting point for understanding the platform’s stated approach. The C2PA specifications also explain that provenance records can be incomplete and do not, by themselves, decide whether content is true.

A fairer approach to artistic provenance

A credible account of an artwork may include photographs of earlier stages, the original source file, layer history, working sketches, physical-material descriptions, dated studio records and the artist’s own account. Any one of these can have limitations. Considered together, they often provide richer evidence of process than the presence or absence of a single digital indicator.

Platforms should make a meaningful distinction between generated, AI-assisted or edited, and human-created work shared in digital form. When automated systems are uncertain, their language should communicate uncertainty. Artists should have a practical opportunity to correct inaccurate or excessively broad labels.

A work that returns: our recursive studio

At ZGenArt.com, making an artwork often means returning to it repeatedly across different tools and surfaces. A painting may begin with watercolor, oil pastel and Indian ink on paper; the physical work is photographed, digitally altered, repainted with a stylus in Infinite Painter, reconsidered with AI-assisted tools and then revised by the human artist again. Some works begin digitally or with an AI-produced image, while others begin entirely by hand. No fixed recipe describes every work.

James A. Swan leads this process in continuing creative collaboration with Janet and other language-model and imaging partners. The studio’s 360° Recurring Hybrid Art System names the exchange, not a machine capable of making every decision. Attribution is most useful when it describes the actual steps and contributions in a specific piece. Calling every finished PNG simply “AI art” overlooks the materials, hands, choices and successive revisions that may have produced it.

ZGenArt.com is an exhibition and an evolving record of that practice, not an online storefront. The .com address does not change its artistic purpose.

An editorial principle for ZGenArt.com

Our studio publishes art made with physical materials, human digital painting and properly credited collaborative processes. We will identify the medium and contributions on their own terms. A photographed oil pastel will remain an oil pastel. A hand-drawn digital work will be described as digital painting. A work involving generative AI will not be disguised as something else.

Good labeling should illuminate the creative process, not erase it. That is the standard artists, audiences and digital platforms should pursue.

Primary references

Meta: Our Approach to Labeling AI-Generated Content and Manipulated Media (originally published 2024; updated 2025).

C2PA: Content Credentials Explainer (specification reference).